def train_data():
dummy_clf = pipelines.bag_of_words(
classifier=DummyClassifier(random_state=0, strategy="stratified"))
log_regression = pipelines.bag_of_words(classifier=LogisticRegression(
random_state=0, n_jobs=-1, max_iter=2000))
log_regression_tfidf = pipelines.bag_of_words(
classifier=LogisticRegression(random_state=0, n_jobs=-1,
max_iter=2000),
tf_idf=True)
linear_svc = pipelines.bag_of_words(
classifier=LinearSVC(max_iter=2000, random_state=0))
linear_svc_tfidf = pipelines.bag_of_words(classifier=LinearSVC(
max_iter=2000, random_state=0),
tf_idf=True)
# svc = pipelines.bag_of_words(
# classifier=SVC(gamma='scale', kernel='rbf')
# )
bow_pipelines = [("Binary: BoW + Dummy", dummy_clf),
("Binary: BoW + LR", log_regression),
("Binary: BoW + LR + TFIDF", log_regression_tfidf),
("Binary: BoW + LinearSVC", linear_svc),
("Binary: BoW + LinearSVC + TFIDF", linear_svc_tfidf)
# ("Binary: BoW + SVC", svc)
]
for name, model in bow_pipelines:
model.set_params(vect__ngram_range=(1, 4))
yield (name, model)
def bag_of_words_pipeline():
log_regression = pipelines.bag_of_words(classifier=LogisticRegression(
C=10.0))
log_regression_tfidf = pipelines.bag_of_words(
classifier=LogisticRegression(C=10.0), tf_idf=True)
bow_pipelines = [("BoW + LR", log_regression),
("BoW + LR + TFIDF", log_regression_tfidf)]
for name, pipe in bow_pipelines:
pipe.set_params(vect__max_features=50000)
yield (name, pipe)
def bag_of_words_pipeline():
log_regression = pipelines.bag_of_words(
classifier=LogisticRegression(random_state=0,
solver='saga',
multi_class='ovr',
n_jobs=-1,
max_iter=2000))
log_regression.set_params(vect__ngram_range=(1, 4))
return log_regression
def gradient_boosting_pipeline():
gradent_boosting = pipelines.bag_of_words(
classifier=GradientBoostingClassifier(
n_estimators=5000, learning_rate=0.2, random_state=10))
gradent_boosting_tfidf = pipelines.bag_of_words(
classifier=GradientBoostingClassifier(n_estimators=5000,
learning_rate=0.2,
random_state=10),
tf_idf=True)
gb_pipelines = [("BoN + GB", gradent_boosting),
("BoN + GB + TFIDF", gradent_boosting_tfidf)]
for name, pipe in gb_pipelines:
pipe.set_params(
vect__ngram_range=(1, 5),
vect__max_features=500000,
)
yield (name, pipe)
def bag_of_ngrams_pipelines():
log_regression = pipelines.bag_of_words(
classifier=LogisticRegression(C=10.0), )
log_regression_tfidf = pipelines.bag_of_words(
classifier=LogisticRegression(C=10.0), tf_idf=True)
linear_svc = pipelines.bag_of_words(classifier=LinearSVC(), )
linear_svc_tfidf = pipelines.bag_of_words(classifier=LinearSVC(),
tf_idf=True)
multinomial_nb = pipelines.bag_of_words(classifier=MultinomialNB(), )
multinomial_nb_tfidf = pipelines.bag_of_words(classifier=MultinomialNB(),
tf_idf=True)
bon_pipelines = [
("BoN + LR", log_regression),
("BoN + LR + TFIDF", log_regression_tfidf),
("BoN + SVC", linear_svc),
("BoN + SVC + TFIDF", linear_svc_tfidf),
("BoN + MNB", multinomial_nb),
("BoN + MNB + TFIDF", multinomial_nb_tfidf),
]
for name, pipe in bon_pipelines:
pipe.set_params(vect__ngram_range=(1, 5), vect__max_features=500000)
yield (name, pipe)
def bag_of_words_pipeline():
log_regression = pipelines.bag_of_words(
classifier=LogisticRegression(random_state=0,
solver='saga',
multi_class='ovr',
n_jobs=-1,
max_iter=2000))
log_regression_tfidf = pipelines.bag_of_words(
classifier=LogisticRegression(random_state=0,
solver='saga',
multi_class='ovr',
n_jobs=-1,
max_iter=2000),
tf_idf=True)
multinomial_nb = pipelines.bag_of_words(classifier=MultinomialNB())
multinomial_nb_tfidf = pipelines.bag_of_words(classifier=MultinomialNB(),
tf_idf=True)
linear_svc = pipelines.bag_of_words(
classifier=LinearSVC(multi_class='ovr', max_iter=2000, random_state=0))
linear_svc_tfidf = pipelines.bag_of_words(classifier=LinearSVC(
multi_class='ovr', max_iter=2000, random_state=0),
tf_idf=True)
# tune: max_depth, min_child_weight, n_estimatorss
xgb = pipelines.bag_of_words(
classifier=XGBClassifier(learning_rate=0.2,
n_estimators=1000,
max_depth=7,
objective='multi:softprob',
n_class=3,
n_jobs=-1,
random_state=0,
min_child_weight=3))
xgb_tfidf = pipelines.bag_of_words(classifier=XGBClassifier(
learning_rate=0.2,
n_estimators=1000,
max_depth=7,
objective='multi:softprob',
n_class=3,
n_jobs=-1,
random_state=0,
min_child_weight=3),
tf_idf=True)
sgd_classifier = pipelines.bag_of_words(
classifier=SGDClassifier(max_iter=2000, n_jobs=-1, random_state=0))
sgd_classifier_tfidf = pipelines.bag_of_words(classifier=SGDClassifier(
max_iter=2000, n_jobs=-1, random_state=0),
tf_idf=True)
bow_pipelines = [("BoW + LR", log_regression),
("BoW + LR + TFIDF", log_regression_tfidf),
("BoW + MNB", multinomial_nb),
("BoW + MNB + TFIDF", multinomial_nb_tfidf),
("BoW + LinearSVC", linear_svc),
("BoW + LinearSVC + TFIDF", linear_svc_tfidf),
("BoW + XGBoost", xgb),
("BoW + XGBoost + TFIDF", xgb_tfidf),
("BoW + SGDClassifier", sgd_classifier),
("BoW + SGDClassifier + TFIDF", sgd_classifier_tfidf)]
for name, model in bow_pipelines:
model.set_params(vect__ngram_range=(1, 4))
yield (name, model)